Construction of a statistical validation system for formulation of a rule well featuring crustal activities

Based on the relation between the occurrence of large inland earthquakes and the spatiotemporal relationship between any two kinds of geophysical measures, we aim to formulate a unified scientifically-testable rule for comprehensive understanding and monitoring of crustal activities accompanied by the occurrence of large inland earthquakes. In order to feature the spatiotemporal relationship and validate its relation to the occurrence of large inland earthquakes, we developed a statistical validation system which leads to formulation of a rule which contributes to monitoring of crustal activities. The system requires the input of observational data at the first step and leads to the output of contingency tables showing the relations between the occurrence of large inland earthquakes and candidates for the rule most appropriately featuring crustal activities, probability gains of the occurrence rates of large inland earthquakes according to the rule candidates, and the error diagram (Molchan diagram) of the monitoring rule showing the highest probability gain at the last step. The system is composed of the following four steps: Step 1: Formulation and updating of a database of geophysical measures from observational data, Step 2: Comparison between any two kinds of geophysical measures, Step 3: Quantification of their spatiotemporal relationship: expression using a statistical index, Step 4: Validation of the relation between the temporal changes in the statistical index defined at step 3 and the occurrence of large inland earthquakes; the output of contingency tables, probability gains of the realization rates of large inland earthquakes according to the candidates for the most useful monitoring rule, and the error diagram (Molchan diagram) of the best rule in terms of probability gains. We introduce an example of application of the system to the database reflecting crustal deformations (GPS data maintained by Geographical Survey Institute, Japan) and seismic activities (hypocenter catalog operated by Japan Meteorological Agency) over the Japanese islands. We defined the lower limit of large inland earthquakes as M≥6 in 2003 to 2008. In this study, we especially notice an evaluation process using the error diagram as well as contingency tables and probability gains.